We really do need fewer (but higher-quality) patents, and we need a more predictable system. With any luck, this will deliver that. And I speak as the inventor of 20+ US Patents. (Corporate ones; an individual inventor might feel otherwise.) :-)

--Greg

Perhaps I'm misunderstanding what happened here, but my impression was that if you quit before the company went public, then they could buy back any of your shares at par. That actually seems pretty reasonable to me. I worked for a place once where a couple of guys who were there for the first year got lumps of stock they in no way deserved, yet they ended up with a nice payout when the company was acquired almost ten years later. A rule that says "you only keep the stock if you're with us when we go public or get acquired" seems very reasonable to me.

What's bad is to have such a rule but hide it from people.

--Greg

Yep, UW's compling program is really, really good. Someone who's into bioinformatics and hates languages probably wouldn't enjoy it much, though.

--Greg

Compiler errors can't be bugs, since the program won't run at all! Or are we talking about bugs *in the compiler?* :-) --Greg

Not that anyone besides us ever got to see those numbers. Anyway, as time went by, it became clear that bot-generated queries often outnumbered real ones anyway. (Another reason not to put too much stock in the query logs.) --Greg

Yeah, there's a reason why we didn't immediately figure it out.

And, yes, people certainly do enter "yahoo" as a search query: it turns out that a lot of people depended on the fact that we'd return a whole pages of results from yahoo.com, and they'd use that as an easier way to navigate the site internals.

However, it's a long step from "people do that sometimes" to "5% of all queries are like that."

When we changed the engine so it only gave two results from a single site, we did get complaints from people who really did type "yahoo" as a query. But it continued to be the #3 result. Another really strong clue that something was rotten.

--Greg

I think "internet explorer" typed out in full should have been the best clue. Turned out it was generated by (if I recall correctly) an MSN application of some kind.

I worked on MSN Search (later "Live Search") so I can answer a few of these for you: 1) There was very little collaboration with the MSN teams. MSN is generally despised at Microsoft, and to get people to come to Search we had to reassure them that it wasn't "really" part of MSN. For their part, the MSN people seemed to try really hard to live up to their "it can't be done" reputation. For example, the MSN team controlled the UI, and even though a top customer complaint was that there wasn't enough space for users to type their queries, no force in the Universe was powerful enough to make the MSN guys widen it. (Their design rules required it be usable by people whose display was a TV set.) 2) Yeah, the MSN data was worthless. First, there wasn't that much of it; rather than saving the raw data, they had a process for computing digests of it, and that's all we could get. Also, that digest process was full of bugs. For example, for years it told us the top queries were "google," "internet explorer" and "yahoo"; it was obvious this was a bug, but our management couldn't get the MSN team to do anything about it. 3) As Yusuf suggests in his article, the cumuative Search and Click data is NOT what you need to produce a good search engine. One of the most frustrating things about working on Search at Microsoft was Management's obsession with head queries. They had several articles of faith that didn't accord with reality, but this was one of the worst. Good news for Microsoft if they've finally figured this out. Of course, almost all the people responsible for the original mess are long gone now. 4) The Google-worship was nauseating. We wasted all kinds of effort trying to duplicate features that obviously didn't work even for Google (news being an obvious example) whereas new features that might have been helpful consistently got killed with "Google doesn't do that." In many cases, this argument was used for technologies where no one had any reasonable clue what Google actually did. --Greg

Often the upper-management guys are non-technical and they're loath to get involved in any sort of dispute over a technical issue. When the dispute is between someone technical and someone non-technical, I'll at least say that I don't think they generally favor the non-technical guy; they just don't do anything.

This doesn't happen all the time, and I can hardly speak for every group at Microsoft either. But it happens often enough in enough groups to be a problem. And it's my perception that it happens more and more.

--Greg

You'd be surprised how many people say that, and yet when the software gets to see their dynamic signature (as points in time, not just an image), it easily finds the things that make their signature unique and clearly distinguishes it from anyone else's. Or so it appeared, anyway. As I said, we never really got to do a proper evaluation, but we did do enough to determine that a lot of our intuitions about the problem were simply wrong.

--Greg

Actually I'd claim that the real problem with Pocket PC and Tablet PC (and I worked on both products) was that they predated cheap, universal wireless communications. The effort to send a text message from a cell phone is much greater than any of the input methods on either of those devices, yet millions of people do it all the time. That wireless network totally changed the value proposition of one of these devices, and before that, they just weren't worth the trouble.

In plain terms, the isolated Tablet was little more than a crippled laptop, and the isolated Pocket PC was almost completely useless. Attach them to a network, though, and they become something magical. Something none of us working on them was wise enough to foresee.

--Greg

In the case of Tablet Computers, it has been a very long wait for someone to make something that worked!

--Greg

For a long time, Microsoft avoided the sort of sclerosis that seemed to affect big companies like IBM, AT&T, and Xerox. People attributed it to things like Microsoft's amazing decentralization of responsibility (in which each VP operates much like the CEO of his or her own startup) to the "Program Manager" role that separated the job of collecting requirements from the job of progamming them. But over the last decade, things seem to have gradually frozen up.

I was at Microsoft at the same time Dick Brass was (and even reported into his organization for a while), so I'm going to beat up on him a little. (He won't mind.) We really wanted Tablet PC to be viable without a keyboard because it made such a difference in weight and size. There are a number of problems with operating such a device that way, but simply logging into it was a bear. Virtual keyboard and handwriting recognition solutions were both miserable, so we looked at biometrics. Now for a Tablet PC, the obvious biometric is signature verification, but one powerful individual in Dick Brass's organization had such a passion for fingerprint verification, that he effectively stopped us from even evaluating signature verification systems. Never mind that the fingerprint systems were extra hardware, stuck out the side, were easy to break off, etc. -- this individual was impervious to reason. Dick could have broken the logjam, but wouldn't get involved. Ultimately, we did nothing, and no serious keyboardless Tablet PCs were ever made (that I know of). This wasn't the only reason, but it was enough by itself.

This pair of problems -- the non-technical guy who kills ideas and can't be reasoned with plus upper management that can't get involved -- seems to have become depressingly common across the whole company. Bright people get discouraged and leave. People who thrive on stifling other people stay.

Where I do disagree with Dick is that I think a VP still has enough autonomy to make his/her own org successful. Microsoft's top management could still fix this problem if it consistently focused on getting and keeping the right VPs and eliminating the bad ones. I think the problem and the solution start and end in the same place.

--Greg